Control method for preparing carbon fiber composite board

Abstract

The invention discloses a control system and method for preparing a carbon fiber composite board. The control method for preparing the carbon fiber composite board comprises the following steps: acquiring melt temperature data acquired by a temperature sensor and an original electrical impedance spectrogram acquired by an impedance analyzer; performing temperature compensation on the original electrical impedance spectrogram based on melt temperature data to obtain a temperature compensation electrical impedance spectrogram; inputting the temperature compensation electrical impedance spectrogram into a trained dispersion state analysis machine learning model to obtain a real-time carbon fiber dispersion index; inputting the real-time carbon fiber dispersity index and a preset target divergence index range into an expert system rule base to obtain a screw rotation speed expected value; and inputting the real-time screw rotating speed value and the screw rotating speed expected value into a PLC (Programmable Logic Controller) to obtain a screw rotating speed control instruction, wherein the screw rotating speed control instruction is used for representing the rotating speed of a screw motor. Therefore, the dispersion uniformity of the carbon fibers in the resin matrix can be monitored and controlled on line.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of carbon fiber composite board preparation, and particularly relates to a control method for carbon fiber composite board preparation. BACKGROUND

[0002] Carbon fiber composite materials have been increasingly widely used in high-end manufacturing fields such as aerospace, automobile industry, rail transportation and sports equipment due to their light weight, high strength, corrosion resistance, fatigue resistance and excellent design freedom.

[0003] As a common carbon fiber reinforced thermoplastic resin composite CFRTP product form, the final performance of the carbon fiber composite board depends largely on the uniformity of the dispersion of carbon fibers in the resin matrix and the interfacial bonding strength between the fibers and the matrix. In the preparation process of the carbon fiber composite board, especially in the step involving melt blending, how to accurately control the process parameters to ensure the effective dispersion of carbon fibers, avoid agglomeration and achieve excellent wetting is a key link to determine the macro mechanical properties of the composite material.

[0004] However, in the existing carbon fiber composite board preparation process, the control of the dispersion state of carbon fibers in the resin melt often faces many challenges. The first problem is the lack of real-time, online monitoring means for the dispersion state of carbon fibers in the melt blending process. Production personnel must take samples after production or intermittently during production, and then perform time-consuming and complex sample preparation and analysis. This hysteresis makes it difficult to adjust the process parameters in a timely and effective manner once the problem of poor dispersion is found, resulting in increased production costs and difficulty in achieving timely and effective adjustment of process parameters.

[0005] Secondly, the existing process control relies mainly on the feedback of indirect parameters or the experience of operators. For example, operators may infer the dispersion inside the melt according to the fluctuations of macroscopic parameters such as the torque of the extruder, the die pressure, the melt temperature, etc. However, these parameters are the result of the combined action of multiple factors (such as material filling degree, screw shear, melt viscosity, etc.), and the correlation between them and the dispersion state of carbon fibers is not direct and unique, with insufficient sensitivity and specificity.

[0006] Furthermore, the melt blending process itself is a complex multi-physical field coupling process involving fluid mechanics, thermodynamics and material science. The existing process often uses fixed process parameter combinations, which are difficult to dynamically adapt to material property fluctuations and disturbances in the processing process, and cannot achieve optimal control of dispersion.

[0007] Therefore, there is a need for a control method for carbon fiber composite board preparation that can improve the uniformity of the dispersion of carbon fibers in the resin matrix. SUMMARY

[0008] An advantage of the present application is to provide a control method for carbon fiber composite plate preparation, wherein the control method for carbon fiber composite plate preparation can monitor and control the dispersion uniformity of carbon fibers in the resin matrix during the preparation process of carbon fiber composite plate.

[0009] According to one aspect of the present application, a control method for carbon fiber composite plate preparation is provided, comprising:

[0010] S1: acquiring melt temperature data collected by a temperature sensor and raw electrical impedance spectrum collected by an impedance analyzer;

[0011] S2: temperature compensation of the raw electrical impedance spectrum based on the melt temperature data to obtain a temperature-compensated electrical impedance spectrum;

[0012] S3: inputting the temperature-compensated electrical impedance spectrum into a trained dispersion state analysis machine learning model to obtain a real-time carbon fiber dispersion index;

[0013] S4: inputting the real-time carbon fiber dispersion index and a preset target dispersion index range into an expert system rule base to obtain a screw speed expected value;

[0014] S5: inputting the real-time screw speed value and the screw speed expected value into a PLC controller to obtain a screw speed control instruction, which is used to represent the screw motor speed.

[0015] In an embodiment of the control method for carbon fiber composite plate preparation according to the present application, the S2 comprises: extracting a first alternating current conductivity real part of a first frequency point from the raw electrical impedance spectrum; extracting a collection point melt temperature corresponding to the first frequency point from the melt temperature data; and performing temperature compensation based on the collection point melt temperature according to the following formula to obtain a compensated first alternating current conductivity real part, wherein the formula is:

[0016] σ' c = σ' m * exp[(E a / K)*(1 / T m -1 / T r )]

[0017] wherein σ' m is the first alternating current conductivity real part, σ' c is the compensated first alternating current conductivity real part, E a is the activation energy, K is the Boltzmann constant, T m is the collection point melt temperature, and T r is the reference temperature.

[0018] In an embodiment of the control method for carbon fiber composite plate preparation according to the present application, the dispersion state analysis machine learning model comprises a convolution layer-based spectrum feature extraction module and a linear output layer-based dispersity index regression module.

[0019] In an embodiment of the control method for carbon fiber composite plate preparation according to the present application, the convolution layer is a hollow convolution layer, and the linear output layer comprises a full connection layer and a decoder.

[0020] In an embodiment of the control method for carbon fiber composite plate preparation according to the present application, the S3 comprises: inputting the temperature-compensated electrical impedance spectrum into the convolution layer-based spectrum feature extraction module to obtain an electrical impedance spectrum distribution feature encoding vector; and inputting the electrical impedance spectrum distribution feature encoding vector into the linear output layer-based dispersity index regression module to obtain the real-time carbon fiber dispersity index.

[0021] In an embodiment of the control method for carbon fiber composite plate preparation according to the present application, after the S3, the temperature-compensated electrical impedance spectrum is inputted into the convolution layer-based spectrum feature extraction module to obtain an electrical impedance spectrum distribution feature encoding vector, the method further comprises: performing spectrum embedding of the electrical impedance spectrum distribution feature encoding vector under the action of a feature subgroup to obtain an electrical impedance spectrum distribution feature spectrum embedding vector; and generating a dynamic spectrum anchor point based on the electrical impedance spectrum distribution feature spectrum embedding vector to obtain a dynamic electrical impedance spectrum distribution feature vector; wherein the electrical impedance spectrum distribution feature encoding vector is inputted into the linear output layer-based dispersity index regression module to obtain the real-time carbon fiber dispersity index, which comprises: inputting the dynamic electrical impedance spectrum distribution feature encoding vector into the linear output layer-based dispersity index regression module to obtain the real-time carbon fiber dispersity index.

[0022] In an embodiment of the control method for carbon fiber composite plate preparation according to the present application, the spectrum embedding of the electrical impedance spectrum distribution feature encoding vector under the action of a feature subgroup to obtain an electrical impedance spectrum distribution feature spectrum embedding vector comprises: spectrum embedding parameters based on the activation energy, the melt temperature of the collection point and the reference temperature; and performing spectrum-conserved trigonometric function symmetry constraint on the electrical impedance spectrum distribution feature encoding vector by the spectrum embedding parameters to obtain a first electrical impedance spectrum distribution feature spectrum embedding vector and a second electrical impedance spectrum distribution feature spectrum embedding vector.

[0023] In an embodiment of the control method for carbon fiber composite board preparation according to the present application, the dynamic spectrum anchor point generation based on the embedding mapping spectrum and the embedding vector of the distribution characteristic spectrum of the electrical impedance spectrum is used to obtain the dynamic distribution characteristic vector of the electrical impedance spectrum, including: performing dynamic spectrum anchor point generation based on the first embedding vector of the distribution characteristic spectrum of the electrical impedance spectrum and the second embedding vector of the distribution characteristic spectrum of the electrical impedance spectrum to obtain the dynamic distribution characteristic vector of the electrical impedance spectrum.

[0024] In an embodiment of the control method for carbon fiber composite board preparation according to the present application, the S5 includes: calculating an error value between the real-time screw speed value and the screw speed expected value; inputting the error value into the PLC controller to obtain a proportional adjustment part, a differential adjustment part and an integral adjustment part; and generating the screw speed control instruction based on the proportional adjustment part, the differential adjustment part and the integral adjustment part.

[0025] According to another aspect of the present application, the present application provides a control system for carbon fiber composite board preparation, which includes:

[0026] A data acquisition unit is configured to acquire melt temperature data collected by a temperature sensor and raw electrical impedance spectrum collected by an impedance analyzer;

[0027] A temperature compensation unit is configured to perform temperature compensation on the raw electrical impedance spectrum based on the melt temperature data to obtain a temperature-compensated electrical impedance spectrum;

[0028] A carbon fiber dispersibility index acquisition unit is configured to input the temperature-compensated electrical impedance spectrum into a trained dispersal state analysis machine learning model to obtain a real-time carbon fiber dispersibility index;

[0029] A screw speed acquisition unit is configured to input the real-time carbon fiber dispersibility index and a preset target dispersibility index range into an expert system rule base to obtain a screw speed expected value;

[0030] A screw control instruction acquisition unit is configured to input a real-time screw speed value and a screw speed expected value into a PLC controller to obtain a screw speed control instruction, which is used to represent a screw motor speed.

[0031] The further purposes and advantages of the present application will be fully apparent from the following description and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0032] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0033] Figure 1 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application.

[0034] Figure 2 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application.

[0035] Figure 3 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application.

[0036] Figure 4 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application.

[0037] Figure 5 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application.

[0038] Figure 6 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application.

[0039] Figure 7 Fig. 1 illustrates a flow chart of a control method for carbon fiber composite board preparation according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein.

[0041] It can be understood that the term "one" should be understood as "at least one" or "one or more", that is, in an embodiment, the number of one element can be one, and in another embodiment, the number of the element can be multiple, and the term "one" cannot be understood as a limitation on the number. "Multiple" refers to greater than or equal to two.

[0042] Although ordinal numbers such as "first", "second", etc. will be used in describing various components, those components are not limited herein. The ordinal numbers are used only to distinguish one component from another component, for example, a first component can be referred to as a second component, and likewise, a second component can also be referred to as a first component, without departing from the teachings of the present application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0043] The terminology used herein is for the purpose of describing various embodiments only and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "has", when used in this specification, specify the presence of stated features, numbers, operations, components, elements, or combinations thereof, but do not preclude the presence or addition of one or more other features, numbers, operations, components, elements, or combinations thereof.

[0044] As mentioned above, in the existing carbon fiber composite plate preparation process, the control of the dispersion state of carbon fibers in the resin melt often faces many challenges. The primary problem is the lack of real-time, online monitoring means for the dispersion state of carbon fibers during the melt blending process. Secondly, the existing process control relies more on the feedback of indirect parameters or the experience of operators. Furthermore, the existing process often uses fixed process parameter combinations, which are difficult to dynamically adapt to material property fluctuations and disturbances in the processing process, and cannot achieve optimal control of dispersion.

[0045] Therefore, there is a need for a control method for carbon fiber composite plate preparation that can improve the uniformity of carbon fiber dispersion in the resin matrix.

[0046] The present application monitors the process parameters affecting carbon fiber dispersion in the high-temperature and high-pressure melt blending process in real time, and intelligently analyzes the process parameters based on a machine learning model to obtain a carbon fiber dispersion index. Then, based on the carbon fiber dispersion index, the key process parameters of the extruder, such as screw speed, are adjusted to achieve closed-loop optimization control of carbon fiber dispersion uniformity, thereby achieving intelligent and adaptive control of carbon fiber dispersion uniformity in the melt blending process, ensuring the high quality and production stability of the composite plate.

[0047] Based on this, the application provides a control method for the preparation of carbon fiber composite plates, which comprises the following steps: S1: acquiring melt temperature data collected by a temperature sensor and an original electric impedance spectrum collected by an impedance analyzer; S2: performing temperature compensation on the original electric impedance spectrum based on the melt temperature data to obtain a temperature-compensated electric impedance spectrum; S3: inputting the temperature-compensated electric impedance spectrum into a trained dispersion state analysis machine learning model to obtain a real-time carbon fiber dispersibility index; S4: inputting the real-time carbon fiber dispersibility index and a preset target dispersibility index range into an expert system rule base to obtain a screw rotation speed expectation value; and S5: inputting a real-time screw rotation speed value and the screw rotation speed expectation value into a PLC controller to obtain a screw rotation speed control instruction, wherein the screw rotation speed control instruction is used to represent a screw motor rotation speed.

[0048] Correspondingly, as shown in Figures 1 to 6 , the control method for the preparation of carbon fiber composite plates according to the embodiments of the application is illustrated. As shown in Figure 1 , the control method for the preparation of carbon fiber composite plates comprises the following steps: S1: acquiring melt temperature data collected by a temperature sensor and an original electric impedance spectrum collected by an impedance analyzer; S2: performing temperature compensation on the original electric impedance spectrum based on the melt temperature data to obtain a temperature-compensated electric impedance spectrum; S3: inputting the temperature-compensated electric impedance spectrum into a trained dispersion state analysis machine learning model to obtain a real-time carbon fiber dispersibility index;

[0049] S4: inputting the real-time carbon fiber dispersibility index and a preset target dispersibility index range into an expert system rule base to obtain a screw rotation speed expectation value; and S5: inputting a real-time screw rotation speed value and the screw rotation speed expectation value into a PLC controller to obtain a screw rotation speed control instruction, wherein the screw rotation speed control instruction is used to represent a screw motor rotation speed.

[0050] Specifically, in step S1, melt temperature data collected by a temperature sensor and an original electric impedance spectrum collected by an impedance analyzer are acquired. Specifically, considering that there is a large difference in electrical conductivity / dielectric constant between carbon fibers and resin matrices, the dispersion state of fibers (from large agglomerates to uniformly dispersed single fibers, and to the conductive network that may be formed) will cause significant changes in the overall electrical response of the melt, and the electric impedance spectrum is used as basic data for judging the dispersibility of carbon fibers in the application.

[0051] The impedance analyzer calculates the complex impedance, complex dielectric constant, conductivity, loss tangent and other parameters at each frequency point according to the original electrical signal, for example, the first complex impedance Z', the first complex dielectric constant ε', the first conductivity σ', and the first loss tangent tanδ' at the first frequency point; the first complex impedance Z", the second complex dielectric constant ε", the second conductivity σ", and the second loss tangent tanδ" at the second frequency point, to form an original electrical impedance spectrum. The original electrical signal is derived from the multi-electrode sensor in the extruder port flow cell.

[0052] It is worth mentioning that the electrical impedance spectrum measures the electrical response of the material under different frequency alternating current fields. For polymer melts and their composites (such as carbon fiber / resin melts), their electrical properties (such as conductivity, dielectric constant, and dielectric loss) are highly sensitive to temperature. During the melt blending process, even if the control system used to prepare the carbon fiber composite plate tries to maintain the set temperature, the actual melt temperature may fluctuate within a small range due to factors such as shear heat generation, feed fluctuation, and environmental changes. If temperature compensation is not performed, the changes in electrical properties caused by these temperature fluctuations will be superimposed on the changes in electrical properties caused by changes in the dispersion state of carbon fibers, making it difficult for the machine learning model to accurately distinguish whether the dispersion has changed or only the temperature has changed, thereby leading to misjudgment of the dispersion.

[0053] The present application proposes to correct the original electrical impedance spectrum measured at different actual melt temperatures to a spectrum at a unified reference temperature. In this way, the compensated spectrum mainly reflects the changes in the microstructure of the material (such as the dispersion state of carbon fibers), rather than the effects of temperature fluctuations.

[0054] Correspondingly, in step S2, the original electrical impedance spectrum is temperature-compensated based on the melt temperature data to obtain a temperature-compensated electrical impedance spectrum. Specifically, first, obtain the temperature parameter-spectrum key parameter mapping relationship function parameters; then, obtain the temperature-compensated electrical impedance spectrum based on the temperature parameter-spectrum key parameter mapping relationship function parameters.

[0055] Specifically, the temperature parameter-spectrum key parameter mapping relationship function parameters are obtained through calibration. First, under the premise of keeping the dispersion state of carbon fibers consistent, collect electrical impedance spectra at different temperatures; then, extract key parameters (for example, such as DC conductivity, relaxation frequency, dielectric constant platform value) from the electrical impedance spectrum to obtain multiple temperature-spectrum key parameter data points, and fit the multiple temperature-spectrum key parameters through a physical model (for example, Arrhenius equation, WLF equation) or an empirical model (for example, polynomial fitting) to obtain the temperature parameter-spectrum key parameter mapping relationship function parameters; and select a reference temperature; wherein the reference temperature is the target set temperature of the process or the center temperature within the calibration data coverage range.

[0056] In one example of the present application, the extruder is set to run at N different stable temperatures, for example: T1 = 190℃ (463.15K), T2 = 200℃ (473.15K), …, TN = 210℃ (483.15K).

[0057] In the temperature parameter-spectrogram key parameter calibration process, after each temperature is stabilized, multiple sets of impedance spectrograms are collected, and the real part σ' of the alternating current conductivity at a certain fixed low frequency point (for example, 1 Hz) is extracted, and the average is obtained:

[0058] σ'_1(T1) = 1.0 x 10 -7 S / cm, σ'_2(T2) = 1.8 x 10 -7 S / cm, …, σ'_N(TN) = 3.0 x 10 - 7 S / cm.

[0059] The temperature-spectrogram key parameter data points include: (463.15, 1.0 x 10 -7 ), (1 / 473.15, 1.8 x 10 -7 ), …, (1 / 483.15, 3.0 x 10 -7 ).

[0060] The temperature-spectrogram key parameter data points are fitted by a linear equation ln(σ') = ln(A) - (Ea / k)*(1 / T), where ln(σ') is the dependent variable, (1 / T) is the independent variable, -(Ea / k) is the slope, and ln(A) is the intercept. The temperature-spectrogram key parameter data points are converted to fitting data points: (1 / 463.15, ln(1.0e -7 )), (1 / 473.15, ln(1.8e -7 )), …, (1 / 483.15, ln(3.0e -7 )). After fitting, the temperature parameter-spectrogram key parameter mapping relationship function parameter m = -7000K (i.e. -Ea / k = -7000, Ea / k = 7000) is obtained. The reference temperature is selected as 200℃ (473.15K).

[0061] According to the above calibration method, the temperature parameter-spectrogram key parameter mapping relationship function parameters at different frequency points (for example, 2Hz, 3Hz, etc.) can be obtained.

[0062] In one embodiment of the present application, as Figure 2As shown, the temperature-compensated electrical impedance spectrogram based on the temperature parameter-spectrogram key parameter mapping relationship function parameter includes: S21, extracting a first AC conductivity real part of a first frequency point from the original electrical impedance spectrogram; S22, extracting a melt temperature of a collection point corresponding to the first frequency point from the melt temperature data; S23, based on the melt temperature of the collection point, temperature compensation is performed according to the following formula to obtain a compensated first AC conductivity real part, wherein the formula is:

[0063] σ' c = σ' m *exp[(E a / K)*(1 / T m -1 / T r )]

[0064] Wherein, σ' m is the first AC conductivity real part, σ' c is the compensated first AC conductivity real part, E a is the activation energy, K is the Boltzmann constant, T m is the melt temperature of the collection point, and T r is the reference temperature.

[0065] In an example of the present application, at 195°C (468.15K), the first AC conductivity real part σ' m is 1.5x10 -7 S / cm; the compensated first AC conductivity real part σ' m is: 1.5x10 -7 S / cm*exp[(468.15K / K)*(1 / 468.15K-1 / 473.15K)]≈1.756x10 -7 S / cm.

[0066] The compensated first AC conductivity real part at 195°C (468.15K) can be compared with the compensated first AC conductivity real part at other temperature conditions to evaluate the change in dispersion.

[0067] Only the temperature compensation process for the first frequency point (for example, 1Hz) is described, and it should be understood that in actual application, the temperature compensation of multiple frequency points can be performed according to the above temperature compensation process, and finally the complete temperature-compensated electrical impedance spectrogram is reconstructed.

[0068] In the above example, the temperature compensation process of the AC conductivity real part spectrogram key parameter is mainly described, and it should be understood that in actual application, other spectrogram key parameters are further temperature-compensated.

[0069] It should also be understood that the trained neural network model can be used for temperature compensation. The trained neural network model can be input with the melt temperature data and the original electrical impedance spectrogram to obtain the temperature-compensated electrical impedance spectrogram.

[0070] In step S3, the temperature-compensated electrical impedance spectrogram is input into the trained machine learning model for analyzing the dispersion state to obtain the real-time carbon fiber dispersion index. Specifically, the temperature-compensated electrical impedance spectrogram contains rich frequency domain information (e.g., the changes of impedance modulus, phase angle, dielectric constant, and loss factor with frequency), which is closely related to the distribution state of carbon fibers in the resin matrix (e.g., the size and number of agglomerates, the degree of formation of fiber network, and the fiber spacing). However, considering that the relationship between these information and the distribution state of carbon fibers in the resin matrix is complex and nonlinear, it is difficult to accurately determine the dispersion from the subtle changes in the spectrogram. Machine learning models are good at automatically learning and extracting useful features from high-dimensional complex data. Therefore, the trained machine learning model for analyzing the dispersion state is used to analyze the temperature-compensated electrical impedance spectrogram to obtain the real-time carbon fiber dispersion index, which provides key input data for subsequent closed-loop control.

[0071] In an embodiment of the present application, the machine learning model for analyzing the dispersion state includes a spectrogram feature extraction module based on a convolutional layer and a dispersion index regression module based on a linear output layer. The spectrogram feature extraction module extracts deep features related to the dispersion state of carbon fibers from the subtle changes in the spectrogram. The dispersion index regression module outputs one or a set of quantitative dispersion indexes based on the deep features related to the dispersion state of carbon fibers, such as a comprehensive dispersion score (e.g., 0-100 points, the higher the score, the better the dispersion), an average or maximum size estimate of key agglomerates, a statistical parameter of fiber distribution uniformity (e.g., standard deviation, coefficient of variation), and a prediction confidence.

[0072] In an embodiment of the present application, the convolutional layer is a dilated convolutional layer, and the linear output layer includes a fully connected layer and a decoder. Accordingly, as shown in FIG. 3, step S3 includes: Figure 3 S31, inputting the temperature-compensated electrical impedance spectrogram into the spectrogram feature extraction module based on the convolutional layer to obtain an electrical impedance spectrogram distribution feature encoding vector; and S32, inputting the electrical impedance spectrogram distribution feature encoding vector into the dispersion index regression module based on the linear output layer to obtain the real-time carbon fiber dispersion index.

[0073] It is worth mentioning that, when the original electrical impedance spectrogram is temperature-compensated based on the melt temperature data, the activation energy E aThe temperature-compensated electrical impedance spectrogram is obtained by coupling the non-linear thermal activation kinetics of the melt temperature relative to the reference temperature, so that when the temperature-compensated electrical impedance spectrogram is input into the convolution layer-based spectrogram feature extraction module, the obtained electrical impedance spectrogram distribution feature encoding vector has a multi-scale dynamic non-linear coupling feature under non-linear phase transition critical activation, thereby making the electrical impedance spectrogram distribution feature encoding vector difficult to converge through the linear output layer, and affecting the accuracy of the obtained real-time carbon fiber dispersibility index.

[0074] Based on this, the electrical impedance spectrogram distribution feature encoding vector can be dynamically optimized, and then input into the linear output layer-based dispersibility index regression module to obtain the real-time carbon fiber dispersibility index.

[0075] Specifically, in an embodiment of the present application, as shown in Figure 4 The step S3 further includes, after step S31, S311, performing spectrogram embedding on the electrical impedance spectrogram distribution feature encoding vector under the action of a feature subgroup to obtain an electrical impedance spectrogram distribution feature spectrogram embedding vector; and S312, generating a dynamic spectrogram anchor point based on the electrical impedance spectrogram distribution feature spectrogram embedding vector to obtain a dynamic electrical impedance spectrogram distribution feature vector. Step S32 includes: S321, inputting the dynamic electrical impedance spectrogram distribution feature encoding vector into the linear output layer-based dispersibility index regression module to obtain the real-time carbon fiber dispersibility index.

[0076] In step S311, the electrical impedance spectrogram distribution feature encoding vector is subjected to spectrogram embedding under the action of a feature subgroup to obtain an electrical impedance spectrogram distribution feature spectrogram embedding vector. Specifically, in an embodiment of the present application, the phase transition critical temperature activation response is taken as a non-linear normalized coupling response to activation energy, and the subgroup is subjected to a spectrum-conserving trigonometric function symmetry constraint on the electrical impedance spectrogram distribution feature encoding vector, so that the convolution encoding satisfies the spectrum-conserving constraint in the multi-scale non-linear dynamic mapping space, thereby realizing joint linear modeling on the basis of spectrum adaptation. Before the subgroup is subjected to the spectrum-conserving trigonometric function symmetry constraint, the spectrum embedding parameters need to be calculated.

[0077] Correspondingly, as shown in Figure 5 Step S311 includes: S3111, calculating the spectrum embedding parameters based on the activation energy, the melt temperature at the collection point, and the reference temperature; and S3112, performing spectrum-conserving trigonometric function symmetry constraint on the electrical impedance spectrogram distribution feature encoding vector through the spectrum embedding parameters to obtain a first electrical impedance spectrogram distribution feature spectrogram embedding vector and a second electrical impedance spectrogram distribution feature spectrogram embedding vector.

[0078] In step S3111, a spectral embedding parameter is calculated by the following formula, wherein the formula is:

[0079]

[0080] In step S3112, the first electrical impedance spectrum distribution feature spectral embedding vector and the second electrical impedance spectrum distribution feature spectral embedding vector are calculated by the following formula, wherein the formula is:

[0081] V' = exp(V ⊙arcsinθ )

[0082] V ” = exp(V ⊙arccosθ )

[0083] wherein V' represents the first electrical impedance spectrum distribution feature spectral embedding vector; V ” represents the second electrical impedance spectrum distribution feature spectral embedding vector; V represents the electrical impedance spectrum distribution feature encoding vector; exp(V ⊙arcsinθ ) represents the arccosine θ power of V; and exp(V ⊙arccosθ ) represents the arccosine θ power of V. The spectral conservation trigonometric function symmetry constraint on the spectral embedding parameter θ makes the electrical impedance spectrum distribution feature encoding vector be projected to two orthogonal (or related) directions defined by θ. It is worth mentioning that, although the electrical impedance spectrum distribution feature encoding vector retains or reorganizes certain key structures or relationships in the original spectral graph information in the new embedding space, the core information related to dispersion is conserved, although the form is changed.

[0084] Correspondingly, step S321 comprises: performing dynamic spectral anchor point generation based on the first electrical impedance spectrum distribution feature spectral embedding vector and the second electrical impedance spectrum distribution feature spectral embedding vector to obtain a dynamic electrical impedance spectrum distribution feature vector. Specifically, the dynamic spectral anchor point generation is performed by the following formula:

[0085]

[0086] wherein V I represents a unit feature vector, ω is a predetermined drift coefficient, i.e., the spectral embedding gradient of the subgroup; V' represents the first electrical impedance spectrum distribution feature spectral embedding vector; V" represents the second electrical impedance spectrum distribution feature spectral embedding vector; and represents Hadamard product; represents subtraction; represents direct addition. The spectral embedding gradient of the subgroup is taken as an anchor drift term, and a dynamic spectral anchor is generated by correspondingly shortening the unit response to achieve adaptive coupling of the feature space under the physical consistency constraint brought by the temperature compensation based on the physical parameters, so as to convert the physical consistency of the multi-scale dynamic nonlinear coupling characteristics of the electrical impedance spectroscopy distribution feature encoding vector into adaptive coupling consistency of the feature space, improve the convergence of the electrical impedance spectroscopy distribution feature encoding vector through the linear output layer, and thus improve the accuracy of the obtained real-time carbon fiber dispersibility index.

[0087] In step S4, the real-time carbon fiber dispersibility index and the preset target dispersibility index range are input into an expert system rule base to obtain a screw speed expectation value. Specifically, the screw speed is one of the key process parameters that affect the dispersion of carbon fibers in the melt. It is directly related to the shear rate and the residence time of the material in the extruder. Experienced engineers can usually judge when and how to adjust the screw speed to improve dispersion according to observed phenomena (even if indirect) or historical data. The expert system can encode this experience and form a plurality of rules to form an expert system rule base. The expert system rule base includes a series of IF-THEN rules, the IF part describes one or more situations, states or facts, in this application, the state of the real-time carbon fiber dispersibility index DI_current and the preset target dispersibility index range. The THEN part describes the conclusion that should be drawn or the action that should be taken when the IF condition is met, in this application, the screw speed expectation value RPM_expected. If the IF part of a rule matches the current input data (i.e. the condition is true), the rule is "triggered".

[0088] In an example of the present application, the range of the real-time carbon fiber dispersibility index DI_current is 0-100, the higher the value of the real-time carbon fiber dispersibility index DI_current, the better. The preset target dispersibility index range of the real-time carbon fiber dispersibility index DI_current is 85-95, i.e. the minimum value of the preset target dispersibility index range DI_target_min = 85; the maximum value of the preset target dispersibility index range DI_target_max = 95. The real-time screw speed value is represented by RPM_current; the minimum screw speed RPM_min = 50 rpm; the maximum screw speed RPM_max = 300 rpm; the small amplitude adjustment amount of the screw speed RPM_step_small = 5 rpm; the medium amplitude adjustment amount of the screw speed RPM_step_medium = 10 rpm; the large amplitude adjustment amount of the screw speed RPM_step_large = 20 rpm.

[0089] Expert system rule base (partial example rules) includes: Rule 1: Poor dispersion, need to increase shear significantly; IF (DI_current < 60), THEN RPM_expected = min(RPM_current + RPM_step_large, RPM_max), in other words, if the dispersion is very poor (below 60), then increase the screw speed significantly to enhance shear, but not exceeding the maximum speed limit; Rule 2: Fair dispersion, need to increase shear moderately; IF (DI_current >= 60) AND (DI_current < DI_target_min - 10) / / i.e. DI < 75; THEN RPM_expected = min(RPM_current + RPM_step_medium, RPM_max), in other words, if the dispersion is fair but not very poor (e.g. between 60 and 74), then increase the screw speed moderately; Rule 3: Dispersion slightly below the lower target, need to increase shear slightly; IF (DI_current >= DI_target_min - 10) AND (DI_current < DI_target_min) / / i.e. 75 <= DI < 85, THEN RPM_expected = min(RPM_current + RPM_step_small, RPM_max), in other words, if the dispersion is only slightly below the lower target, then increase the screw speed slightly; Rule 4: Dispersion within the target range, keep current speed; IF (DI_current >= DI_target_min) AND (DI_current <= DI_target_max) / / i.e. 85 <= DI <= 95, THEN RPM_expected = RPM_current, in other words, if the dispersion is within the ideal range, then keep the current screw speed unchanged; Rule 5: Dispersion slightly above the upper target (possibly over-shearing or excessive energy consumption), need to decrease shear slightly; IF (DI_current > DI_target_max) AND (DI_current <= DI_target_max + 5) / / i.e. 95 < DI <= 100, THEN RPM_expected = max(RPM_current - RPM_step_small, RPM_min), in other words, if the dispersion is slightly above the upper target, it can mean that the shear is slightly excessive or unnecessary energy consumption, it can be tried to decrease the speed slightly, but not below the minimum speed limit.Rule 6: Dispersion is much higher than the upper limit of the target (possible serious shear leading to fiber damage), need to reduce the shear of medium amplitude, IF (DI_current> DI_target_max+5) / / that is DI> 100 (assuming the maximum value of DI may exceed 100, or here is another threshold), THEN RPM_expected = max (RPM_current-RPM_step_medium, RPM_min), in other words, if the dispersion is far beyond the target, it may cause excessive fiber breakage, and the speed needs to be reduced more obviously; Rule 7: Dispersion is within the target but continues to decline, preventively increase the shear; IF (DI_current>=DI_target_min) AND (DI_current<=DI_target_max) AND (DI_trend==DECREASING_FAST), THEN RPM_expected = min (RPM_current+RPM_step_small, RPM_max), in other words, even if the current dispersion is qualified, but if it is rapidly declining, it is preventively increased in small amplitude to prevent further deterioration.

[0090] When the input data is real-time carbon fiber dispersion index DI_current = 72, the preset target dispersion index range is 85-95, that is, the minimum value of the preset target dispersion index range DI_target_min = 85; The maximum value of the preset target dispersion index range DI_target_max = 95, the real-time screw speed value RPM_current = 150 rpm; Trigger the THEN part of rule 2: RPM_expected = min (150+10, 300) = min (160, 300) = 160 rpm, that is, the screw speed expected value RPM_expected = 160 rpm.

[0091] In step S5, the real-time screw speed value and the expected screw speed value are input to the PLC controller to obtain a screw speed control command, which represents the screw motor speed. Specifically, after inputting the real-time screw speed value and the expected screw speed value into the Programmable Logic Controller (PLC), the PLC controller first compares the real-time screw speed value and the expected screw speed value and calculates the difference between them (i.e., the error of the real-time screw speed value relative to the expected screw speed value); then, it calculates a suitable control output using a proportional-integral-derivative algorithm. The proportional (P) output is proportional to the current error; the larger the error, the greater the corrective action. The integral (I) output considers the accumulation of error over time, which helps eliminate steady-state error; otherwise, a small persistent error may remain. The derivative (D) output considers the rate of change of the error, which helps suppress oscillations and predict future errors, thereby achieving faster stabilization.

[0092] Accordingly, such as Figure 6 As shown, step S5 includes: S51, calculating the error value between the real-time screw speed value and the expected screw speed value; S52, inputting the error value into the PLC controller to obtain the proportional control part, the derivative control part, and the integral control part; S53, generating the screw speed control command based on the proportional control part, the derivative control part, and the integral control part.

[0093] In one example of the present application, the control period of the PLC is dT=0.1 second, i.e. the PLC performs calculation and output update every 0.1 second, the proportional gain Kp=0.03; the integral gain Ki=0.005; the differential gain Kd=0.005; the control output value is calculated by the following formula: Control_Command_V=SP_V+(Kp*Current_Error)+(Ki*Integral_Error_Sum)+(Kd*Derivative_Error); wherein, Control_Command_V represents the control output value; SP_V represents the voltage value corresponding to the screw speed expected value; Kp represents the proportional gain; Ki represents the integral gain; Kd represents the differential gain; Current_Error represents the current error, which is equal to the error value between the real-time screw speed value and the screw speed expected value; Integral_Error_Sum represents the cumulative error, which is equal to the product of the current error and the control period and the sum of the previous cumulative error; the initial value of the cumulative error is 0; Derivative_Error represents the relative error, which is the ratio of the difference between the current error and the previous error and the control period; Current_Error=RPM_expected-RPM_current; Integral_Error_Sum=Integral_Error_Sum_previous+Current_Error*dT; Derivative_Error=(Current_Error-Previous_Error) / dT.

[0094] In the first control cycle, SP_V = 5.0V; in step S51, the error value between the real-time screw speed value and the screw speed expected value is calculated; wherein, Current_Error = 250RPM-245RPM = 5RPM. In step S52, the error value is input into the PLC controller to obtain the proportional adjustment part, the differential adjustment part and the integral adjustment part; wherein, the proportional adjustment part is Kp*Current_Error = 0.03V / RPM*5RPM = 0.15V; the differential adjustment part is Ki*Integral_Error_Sum = 0.005V / (RPM·s)*(0+5RPM*0.1s = 0.0025V; the integral adjustment part is Kd*Derivative_Error = 0.005V / (RPM / s)*(5RPM-5RPM) / 0.1s = 0RPM / s. In step S53, based on the proportional adjustment part, the differential adjustment part and the integral adjustment part, the screw speed control instruction is generated; wherein, the control output value is 5.0V+0.15V+0.0025V+0V = 5.1525V. Accordingly, the control instruction is obtained: adjust the voltage of the motor driver connected to the extruder motor to 5.1525V, so that the screw speed of the extruder motor is adjusted to about 250RPM.

[0095] The application also provides a control system for preparing a carbon fiber composite plate. Specifically, as shown in Figure 7 The control system for preparing a carbon fiber composite plate includes a data acquisition unit 11, a temperature compensation unit 12, a carbon fiber dispersibility index acquisition unit 13, a screw speed acquisition unit 14 and a screw control instruction acquisition unit 15. The data acquisition unit 11 is used to acquire melt temperature data collected by a temperature sensor and raw electrical impedance spectrogram collected by an impedance analyzer; the temperature compensation unit 12 is used to perform temperature compensation on the raw electrical impedance spectrogram based on the melt temperature data to obtain a temperature-compensated electrical impedance spectrogram; the carbon fiber dispersibility index acquisition unit 13 is used to input the temperature-compensated electrical impedance spectrogram into a trained dispersal state analysis machine learning model to obtain a real-time carbon fiber dispersibility index; the screw speed acquisition unit 14 is used to input the real-time carbon fiber dispersibility index and a preset target dispersibility index range into an expert system rule base to obtain a screw speed expected value; and the screw control instruction acquisition unit 15 is used to input a real-time screw speed value and the screw speed expected value into a PLC controller to obtain a screw speed control instruction, which is used to represent the screw motor speed.

[0096] In summary, the control system and method for preparing carbon fiber composite plate are illustrated. The control method for preparing carbon fiber composite plate can monitor the dispersion uniformity of carbon fiber in the resin matrix in the preparation process of carbon fiber composite plate online and control the dispersion uniformity of carbon fiber in the resin matrix in the preparation process of carbon fiber composite plate by controlling the rotation speed of the screw motor.

[0097] The above description of the application and its embodiments is not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. In summary, if a person skilled in the art is inspired by it, without departing from the purpose of creating the application, without creating a similar structure and embodiment of the technical solution, which should belong to the protection scope of the application.

Claims

1. A control method for carbon fiber composite panel preparation, characterized by, The method comprises the following steps: S1: acquiring melt temperature data collected by a temperature sensor and raw electrical impedance spectrum collected by an impedance analyzer; S2: temperature compensating the raw electrical impedance spectrum based on the melt temperature data to obtain a temperature-compensated electrical impedance spectrum; S3: inputting the temperature-compensated electrical impedance spectrum into a trained dispersion state analysis machine learning model to obtain a real-time carbon fiber dispersibility index; S4: inputting the real-time carbon fiber dispersibility index and a preset target dispersibility index range into an expert system rule base to obtain a screw speed expectation value; S5: inputting a real-time screw speed value and the screw speed expectation value into a PLC controller to obtain a screw speed control instruction, which is used to indicate a screw motor speed.

2. The control method for carbon fiber composite board preparation according to claim 1, characterized by, The S2 comprises the following steps: extracting a first alternating current conductivity real part of a first frequency point from the raw electrical impedance spectrum; extracting a collection point melt temperature corresponding to the first frequency point from the melt temperature data; temperature compensating the collection point melt temperature based on the following formula to obtain a compensated first alternating current conductivity real part, wherein the formula is: σ' c = σ' m * exp[(E a / K)*(1 / T m -1 / T r )] wherein σ' m is the first alternating electrical conductivity real part, σ' c is the compensated first alternating electrical conductivity real part, E a is the activation energy, K is the Boltzmann constant, T m is the melt temperature at the collection point and T r is the reference temperature.

3. The control method for carbon fiber composite board preparation according to claim 1, characterized by, The dispersion state analysis machine learning model comprises a spectrum feature extraction module based on a convolution layer and a dispersibility index regression module based on a linear output layer.

4. The control method for carbon fiber composite board preparation according to claim 3, characterized by, The convolution layer is a hollow convolution layer, and the linear output layer comprises a full connection layer and a decoder.

5. The control method for carbon fiber composite board preparation according to claim 2, characterized by, The S3 comprises the following steps: inputting the temperature-compensated electrical impedance spectrum into the spectrum feature extraction module based on the convolution layer to obtain an electrical impedance spectrum distribution feature encoding vector; inputting the electrical impedance spectrum distribution feature encoding vector into the dispersibility index regression module based on the linear output layer to obtain the real-time carbon fiber dispersibility index.

6. The control method for carbon fiber composite board preparation according to claim 5, characterized by, The S3 further comprises the following steps after inputting the temperature-compensated electrical impedance spectrum into the spectrum feature extraction module based on the convolution layer to obtain an electrical impedance spectrum distribution feature encoding vector: performing spectrum embedding on the electrical impedance spectrum distribution feature encoding vector under the action of a feature subgroup to obtain an electrical impedance spectrum distribution feature spectrum embedding vector; generating a dynamic spectrum anchor point based on the electrical impedance spectrum distribution feature spectrum embedding vector to obtain a dynamic electrical impedance spectrum distribution feature vector; wherein inputting the electrical impedance spectrum distribution feature encoding vector into the dispersibility index regression module based on the linear output layer to obtain the real-time carbon fiber dispersibility index comprises: inputting the dynamic electrical impedance spectrum distribution feature encoding vector into the dispersibility index regression module based on the linear output layer to obtain the real-time carbon fiber dispersibility index.

7. The control method for carbon fiber composite board preparation according to claim 6, characterized by, Performing spectrum embedding on the electrical impedance spectrum distribution feature encoding vector under the action of a feature subgroup to obtain an electrical impedance spectrum distribution feature spectrum embedding vector comprises: a spectrum embedding parameter based on the activation energy, the collection point melt temperature and the reference temperature; performing spectrum-conserving trigonometric function symmetry constraint on the electrical impedance spectrum distribution feature encoding vector through the spectrum embedding parameter to obtain a first electrical impedance spectrum distribution feature spectrum embedding vector and a second electrical impedance spectrum distribution feature spectrum embedding vector.

8. The control method for carbon fiber composite board preparation according to claim 7, characterized by, Generating a dynamic spectrum anchor point based on an embedding mapping spectrum and the electrical impedance spectrum distribution feature spectrum embedding vector to obtain a dynamic electrical impedance spectrum distribution feature vector comprises: The dynamic spectrum anchor point generation is performed based on the first electrical impedance spectrum distribution feature spectrum embedding vector and the second electrical impedance spectrum distribution feature spectrum embedding vector to obtain a dynamic electrical impedance spectrum distribution feature vector.

9. The control method for carbon fiber composite board preparation according to claim 1, characterized by, The S5 comprises: calculating an error value between the real-time screw rotation speed value and the screw rotation speed expected value; inputting the error value into the PLC controller to obtain a proportional adjustment part, a differential adjustment part and an integral adjustment part; generating the screw rotation speed control instruction based on the proportional adjustment part, the differential adjustment part and the integral adjustment part.

10. A control system for carbon fiber composite panel production, characterized by, comprise: a data acquisition unit configured to acquire melt temperature data collected by a temperature sensor and an original electrical impedance spectrum diagram collected by an impedance analyzer; a temperature compensation unit configured to perform temperature compensation on the original electrical impedance spectrum diagram based on the melt temperature data to obtain a temperature-compensated electrical impedance spectrum diagram; a carbon fiber dispersibility index acquisition unit configured to input the temperature-compensated electrical impedance spectrum diagram into a trained dispersion state analysis machine learning model to obtain a real-time carbon fiber dispersibility index; a screw rotation speed acquisition unit configured to input the real-time carbon fiber dispersibility index and a preset target dispersibility index range into an expert system rule base to obtain a screw rotation speed expected value; a screw control instruction acquisition unit configured to input a real-time screw rotation speed value and a screw rotation speed expected value into a PLC controller to obtain a screw rotation speed control instruction, wherein the screw rotation speed control instruction is used to represent a screw motor rotation speed.

Images